ModelQuantiser Class Reference

Public Member Functions
	__init__ (self, weights_file, representative_dataset, representative_example_nr)
	quantise_model (self, output_path)

Public Attributes
	weights_file = weights_file
	representative_dataset = representative_dataset
	representative_example_nr = representative_example_nr
	model = self._load_model()
	input_shape = self.model.input_shape[1:]
	data_generator

Protected Member Functions
	_load_model (self)
	_test_quantised_model (self, tflite_quant_model)

Detailed Description

A class to handle the quantization of a Keras model to TFLite with INT8 precision.

Attributes:
    model (tf.keras.Model): The model to be quantised.
    representative_dataset (str): Path to the representative dataset for quantization.

Constructor & Destructor Documentation

__init__()

__init__	(		self,
			weights_file,
			representative_dataset,
			representative_example_nr )

Initialize the ModelQuantiser.

Args:
    weights_file (str): Path to the Keras model weights file.
    representative_dataset (str): Directory containing the representative dataset.

Member Function Documentation

_load_model()

_load_model ( self )

protected

Load the Keras model from the weights file.

Returns:
    tf.keras.Model: Loaded model.

_test_quantised_model()

_test_quantised_model ( self, tflite_quant_model )

protected

Test the quantised TFLite model by loading it into a TFLite interpreter.

Args:
    tflite_quant_model (bytes): The quantised TFLite model.

quantise_model()

quantise_model	(		self,
			output_path )

Quantise the model to TFLite format with INT8 precision.

Args:
    output_path (str): Path to save the quantised TFLite model.

Member Data Documentation

data_generator

data_generator

Initial value:

=  CustomDataGenerator(
            data_directory=[self.representative_dataset],
            batch_size=1,
            input_shape=self.input_shape,
            stop_training_flag=False,
            shuffle=True
        )

input_shape

input_shape = self.model.input_shape[1:]

model

model = self._load_model()

representative_dataset

representative_dataset = representative_dataset

representative_example_nr

representative_example_nr = representative_example_nr

weights_file

weights_file = weights_file

---

The documentation for this class was generated from the following file:

• model_quantiser.py